1. Two Papers on Network Visualization CPSC 533c Presented by:Jeremy Hilliker 2005-11-07

2. 3D Geographic Network Displays Cox, Eick, He Bell Laboratories 1996

3. Motivation Computer networks can be represented as graphs Often, there is geographic data associated with the network (physical locations) We can put these graphs on a map! But, our ability to extract data from large datasets has not kept pace with our ability to create and gather the data

4. Motivation The telecom dataset is huge! Node-link diagrams do not scale They become overwhelmed, cluttered, and confused  Too many nodes  Too many edges  Edge crossings  Bleh!

5. Motivation We could use graph layout algorithms But then we loose all of the geographic encoding … that stuff was important for easy understanding The paper proposes five solutions which preserve geographic layout by using 3D

6. Why 3D? If we draw arcs instead of lines for edges, we can use pre-attentive depth perception and continuity detection to eliminate the perception of line crossings in the graph

7. Global Networks Position nodes geographically on a globe Draw edges as arcs between them

8. Global Networks Retains spatial information “Eliminates crossings” … doubtful Nodes represented by glyph which can use all of that glyph encoding stuff Arcs encoded by colour for extra info Illuminated by a fixed light which can indicate passage of time… not convinced  What happens at “night”? User restricted to rotations, so can’t get lost

9. Global Networks We can’t see around the globe, so we need a translucency control to see through it  But it’s still confusing if there is edge occlusion That edge clutter is still there  We can filter, losing context  Or we can select how to re-rout an edge Perhaps underground? … it gave great geographic context, but still had all of the 2D layout problems… but worse… I don’t think it helped much

10. Arc Maps Idea: embed a 2D map in 3D space, run edges as arcs in 3D

11. Arc Map Not restricted to a global view, can be of a small region  Leads to “drill-down” = details on demand Arcs in 3D reduce edge clutter  Really get continuation  Can rotate and zoom to get depth perception Arc height can give another encoding of info Can make arcs translucent to reduce occlusion

12. Spoke View Colour code spokes for edge data Colour and size code nodes Nodes can be placed in geographic position if we put the root at the centre of a polar projection  This would make it a filtered 2D global view… But this won’t scale All lines become same length wasting screen space Statement: we can rebuild it using 3D!

13. Helix View What if the spoke view was a top down view of a helix structure? We could rotate it to see everything

14. Pincushion View Arrange nodes on surface of a sphere Lines maintain the same spatial length (radius), but different screen length But nodes are evenly spread out Still need to rotate it to see everything

15. Discussion

16. Visualizing Large- Scale Telecom Nets and Services Koutsofios et al. AT&T Labs 1999

17. Motivation Still have lots of data … lots of small data Old databases don’t handle lots of real-time, small, inter-related data well Understanding full scale of data is needed to manage effectively Goals:  Go from data to business decisions quickly  Raise level of abstraction… lines, not devices  Real-time responsiveness Main contribution: stream based, not query/response

18. Visualization stuff Linked 2D and 3D views (detail + overview) Automated context-preserving pan + zoom Different overlays for different data Semantic zoom (value per state vs. county) Animation over time Can browse and drill-down … seems pretty okay

19. Visualization Stuff

20. Architecture 3 modules  Data collector (and storage)  Aggregator (data processing and pre-proc.)  Visualization (not the important part here) Communication over self-describing data- independent formats  Sounds like a bad idea… (was 1 year after XML)  North-American telecom is a dinosaur Uses advanced systems stuff for fast communication

21. Data Collector Data is converted to the native format Some data has to be aggregated and joined over diverse and content-dependent sources  Mostly because telecom data is a mess Data that is in the right format just needs a schema attached  Doesn’t sound convincing

22. Data Processing Based on pipeline model Concurrent processes are piped together Pipes can:  “Tee”  Filter  Count  Split Pipelines are parallelizable, modular, and simple… fast efficient, and maintainable Pipeline modules are compiled and dynamically linked

23. Data Visualization Interaction pattern:  View data  Focus on something interesting  Query for more details  Re-aggregate and view results Does this by maintaining a link between raw data, aggregate data, and visualized representation

24. Architecture for Performance Does systems stuff to make things fast  Pipelines  Random access files with version stamps  Direct I/O  Memory mapping  Dynamic linking of runtime generated code

25. Discussion

26. Sources Images taken from original papers or found through goolge image search 3D Geographic Network Displays Kenneth C. Cox, Stephen G. Eick, Taosong He. ACM SIGMOD Record Volume 25, Number 4, pp 50-54, 1996 3D Geographic Network Displays Visualizing Large-Scale Telecommunication Networks and Services Eleftherios Koutsofios, Stephen C. North, Russell Truscott, Daniel A. Keim. Proc IEEE Visualization 1999, pp 457-461. Visualizing Large-Scale Telecommunication Networks and Services